Kurt Christian Tubera

Machine learning (ML) models are increasingly used to make decisions in medicine, economics, and other high-stakes environments that demand epistemic trust. The role they play is epistemic because they produce outputs that inform us, and such outputs can either lead to true or false beliefs. Even if most algorithms tend to generate true outputs during their training phase, they are nonetheless untrustworthy because they occasionally produce false outputs when deployed in real-world environments. Philosophers have proposed algorithmic robustness as a response to this problem. Algorithmic robustness requires ML models to be stably truth-conducive, avoiding accidentally true outputs. One leading approach explains algorithmic robustness in terms of modal conditions, which require algorithm outputs to be true in nearby possible worlds for a system to be robust. This paper challenges the modal conditions for algorithmic robustness. I claim that such conditions are unnecessary for robustness. In support of this claim, I argue that the modal conditions for algorithmic robustness fail to preserve constitutive epistemic links (i.e., relations between evidence, belief, and truth) that occur in actual situations. By presenting cases wherein these links obtain without satisfying the modal conditions, I show that an algorithm can remain robust even without invoking counterfactual dependence.

Deathbots are artificial intelligence tools that enable the possibility of encountering deceased loved ones. The advancement of Large Language Models in deathbots makes the imitation of the deceased possible by inputting personal data, messages, photos, and videos. Such AI systems can process new information and provide appropriate responses while mimicking the dead, offering the bereaved a way to sustain connections with them. However, we claim that deathbots put at risk a special kind of knowledge that we can only acquire from grief. In this paper, we hold that grief provides irreplaceable moral insight, illuminating the unique bond between individuals and fostering an appreciation of human goodness. By offering an artificial replacement for this natural process, deathbots may inadvertently prevent grievers from engaging with the true depth of loss and the moral reflection that it inspires. This paper explores how deathbots reshape the grieving experience, potentially distorting the moral dimensions of loss and deterring grievers from the introspection that comes with grieving. We argue that deathbots potentially deprive grievers of having the sort of transformative moral knowledge that arises from the experience of coming to terms with losing someone in its full, unmediated reality.

Medical populism, a performance-based political style that orchestrates antagonistic relations between the people and medical establishments, has served as a concept for analyzing the politicization of healthcare. With its four fundamental features, that is, invoking knowledge claims contrary to medical experts’ advice, simplifying discourse, dramatizing responses to public health crises, and forging divisions, medical populism compromises our access to accurate medical or public health information. Populist leaders abuse their epistemic authority and betray public trust when they propagate false information, medical or otherwise. And insofar as accurate medical or public health information is an epistemic good that we have the right to access, and not having so harms us precisely as knowers, we claim that medical populism violates our epistemic rights.

In ‘The Case of Patient Smith: Pain-Belief, Epistemic Luck, and Acquaintance,’ Elliott Crozat challenged the infallibility of the belief that “I feel pain” by providing a Gettier-type example that shows that such a pain-belief can be fallibly justified and luckily true. We claim that this move is problematic given that the case is not the Gettier sort. To demonstrate this, we first question the causal relation or lack thereof between the subject’s pain-belief and the pain he felt. We argue that this leads us to a dilemma. Either the painbelief and the pain are causally connected or not. If there is a causal connection, the subject’s true belief is justified, not epistemically lucky, and therefore qualifies as knowledge. If we grant no causal connection, we show that this leads to either an unjustified pain-belief or a justified false one. The result of both horns shows that Crozat’s Gettier-type example does not qualify as a bona fide Gettier case.